add a positive list for referencedb-ambiguities to speed things up

[jvollme]

a positive list, containing the exact version of the reference -DB used along with all entires that were preiously assessed as "not contaminations" based on that database should be included also.

This would avoid repeating the same blast analyses with the same exact result over and over again and speed things up.

The list may get rather long over time, therefore:
Test how much the RAM usage increases for huge sets of accessions
Test how much slower a direct search in the text files via binary search tree would be